How five recovery steps compound
- Start: 100% of target product
- Step 1: 85.0% remains
- Step 2: 72.3% remains
- Step 3: 61.4% remains
- Step 4: 52.2% remains
- Step 5: 44.4% remains
Illustrative scenario: a fermentation team models its saleable output using an upstream titre of 45 g/L. It then checks a five-step recovery train. The figures and fundraising setting are hypothetical.
Then a process engineer - contracted, not full-time, brought in two weeks before the raise - ran the downstream numbers properly for the first time.
The purification train required five unit operations: centrifugation, microfiltration, ultrafiltration/diafiltration, a polishing step, and spray drying. Step recovery at each stage was estimated at 82-88%. The team had never measured it across all five steps in sequence.
Illustrative calculation: five sequential steps, each retaining 85 percent of the incoming target product, give 0.85⁵ = 44.37 percent overall recovery.
Starting from 45 g/L of target product in the original broth, that recovery leaves approximately 20.0 g per original litre as recovered target product. This is a mass-balance equivalent, not the concentration of the final purified stream. Calculate saleable output from the complete recovery train.
The assumption to test
The assumption to test is that the strain alone establishes a durable commercial advantage.
A pitch can explain a strain and target molecule without explaining the full recovery train. Make both parts visible in the development and cost model.
Centrifuges, membranes, filtration and drying affect recovery, safety, functionality and costs. Their technical and IP contribution needs examination alongside the organism.
A defensible cost model includes the complete process rather than relying on the appeal of one scientific result.
One-minute brief
For founders: identify where upstream measurements end and downstream assumptions begin. Record what has been measured and what remains to be tested.
For investors: review titre, productivity, overall recovery and the cost boundary together. No one variable establishes the whole operating model.
For R&D teams: develop the strain, product specification and recovery train together. Compare measured constraints before deciding where to invest more work.
What the evidence needs to show
The quantitative reality behind the purification curtain
The economic importance of downstream processing varies by product and process. Start with a defined mass balance and the evidence behind each input.
Downstream costs can be material, but their share varies with product architecture, purity, recovery, equipment and the model boundary. Do not use a universal 50-85 percent share for all food proteins.
GFI’s techno-economic review compares models with different products and assumptions. Feedstock, capital and process performance all matter. Recovery must be modelled for the actual process; the report does not establish one universally dominant variable.
In a hypothetical five-step train, 85 percent recovery at every step gives 44.37 percent overall recovery. At 90 percent per step, overall recovery is 59.05 percent; at 80 percent, it is 32.77 percent. Multiply the fractions. Neither five steps nor these stage yields are universal production benchmarks.
Use equipment and engineering estimates with their scale, product and boundary. A representative upstream estimate is not a quotation for the full facility or its downstream train.
The economic question is the cost per kilogram of saleable product meeting specification. That requires the upstream mass balance, downstream losses and all relevant operating and capital costs.
Three reasons to examine downstream processing
The following three considerations help explain why downstream design belongs in a review of performance, cost and defensibility.
Upstream capabilities can be developed internally or with partners. Assess how difficult the actual organism, pathway, process and product are to reproduce. Available strain-engineering services do not establish that every strain is a commodity or remove the need to examine IP and know-how.
Set purification requirements from safety, regulatory conditions, customer specification and intended functionality together. Food ingredients still need appropriate control of impurities, allergens and host-derived material. The functional target does not justify ignoring safety limits.
Melting, stretching or foaming requirements can help define the target ingredient specification. They do not disclose a named company’s purification process, safety assessment, unit-operation count or achieved cost.
Force 3: The opacity and durability of downstream knowledge. A strain can be sequenced. A pathway can be published. A promoter can be patented and then designed around. A downstream process - the specific sequence of unit operations, the specific operating parameters at each step, the specific fouling patterns on specific membranes, the specific failure modes at specific scales, the specific cleaning protocols that prevent specific contamination events - is institutional knowledge. It accumulates through operational experience, not through research. It lives in batch records, in deviation reports, in the engineering muscle memory of the team that has run the process hundreds of times and knows what breaks at 10,000L that does not break at 200L.
Downstream know-how can be protected and shared in several ways, including patents, trade secrets and technical transfer. Its defensibility depends on the process and available protection; it is not categorically unpatentable or impossible to reproduce.
Related analysis: How to choose a fermentation production site
·31 MarTwo product architectures, two outcomes
The pattern that fails: upstream narrative, downstream assumption
Illustrative failure pattern: a model assumes recovery before the whole train has been measured. A later integrated run shows a lower value. The team must revise its mass balance and cost model. The size and frequency of such gaps are not quantified here.
Lower recovery changes saleable output and the costs allocated to it. Update the mass balance and each relevant cost line, then test sensitivity. A reduction in recovery does not automatically double every component of COGS or establish a particular fundraising outcome.
Cultivated biomass and secreted fermentation proteins involve different processes. A regulatory milestone does not prove production economics, but that distinction alone does not establish the cause of a named company’s financial difficulties.
The pattern that survives: downstream as a design constraint from Day 1
Biomass-as-product and secreted-protein architectures have different recovery requirements. Biomass products still require harvesting, safety controls and often substantial downstream work. Announced plant capacity is not achieved output and does not establish a universal cost advantage.
Vivici and Liberation Labs announced a manufacturing partnership in May 2025. The public announcement provides an example of working with an infrastructure partner; it does not disclose all recovery data or prove the performance of a specific purification train.
The pattern across the survivors: they treated downstream as a design constraint that shaped the product, the host, the expression strategy, and the partnership structure from the earliest stage. They did not treat it as an engineering problem to be solved after the biology was right.
For investors
Ask how the target molecule becomes a saleable ingredient, with the input/output basis and specification made explicit.
Five questions that surface whether the company’s moat is real or assumed:
One. ‘What percentage of your COGS sits in downstream processing, and how did you derive that number?’
A recovery estimate is more useful when it identifies the unit operations, measured losses and relevant process boundary. A percentage copied from another process needs to be labelled as an assumption.
Two. ‘What is your overall DSP recovery from harvest to final product, measured across multiple runs?’
The word ‘measured’ is load-bearing. An assumed recovery rate is not a measurement. A measurement requires defined input weights, defined output weights, and repetition across independent runs.
Three. ‘How many unit operations are in your downstream train, and which ones are you actively working to eliminate?’
Each unit operation has a purpose and an effect on cost and recovery. Assess whether it can be simplified while retaining required safety, functionality and specification. Reducing the count alone does not prove an economic improvement.
Four. ‘What purity specification does your customer actually require, and how does that compare with what your current process produces?’
Check whether each separation step is needed for safety, specification or functionality. Removing an unnecessary step may help economics. Removing a necessary control may make the product unacceptable.
Five. ‘If your titre doubled tomorrow, what percentage of your COGS problem would remain unsolved?’
Compare upstream improvements with downstream improvements in the same model. Their effect depends on cost structure, recovery and product requirements. Doubling titre does not automatically halve total COGS.
The valuation implication: a company with a strong upstream story and an uncharacterised downstream process is not a de-risked investment at Series A or B. It is a company with a narrative and an unmeasured liability. The downstream characterisation should be a pre-investment milestone, not a post-investment discovery.
For R&D teams
Available strain-engineering and fermentation-development services give a team options. Compare those services with the specific process, IP position and technical constraints instead of assuming that every upstream system is easy to reproduce.
Downstream performance can depend on equipment, material interactions, operating windows and accumulated know-how. Record what can be documented, transferred or protected, and what still needs testing.
Three R&D priorities that build the actual moat:
First: design for purification, not purification of whatever you happen to design.
Where biologically and technically appropriate, secretion may avoid a cell-lysis step. Compare expression architecture, host, product quality, safety and recovery before choosing it. A process-design option is not a universal recommendation for every molecule.
Second: characterise recovery across the integrated train and update the cost model as measurements become available.
Keep assumptions visible in an early model. Record their basis and uncertainty, then replace them with relevant measurements when possible. Defined inputs, outputs and repeated runs support a more useful recovery estimate.
Third: define specification using safety, regulatory, customer and functional requirements together.
A functional requirement such as emulsification at a stated pH is only one part of the specification. It does not establish safe impurity limits or remove regulatory conditions. Keep required controls, and investigate whether any additional purification stage adds useful performance.
For founders
Map the connection between the strain, fermentation conditions and final ingredient specification. Show how the measured process and remaining assumptions support the cost estimate.
The downstream process is not a later-stage problem. It is the problem that determines whether your upstream performance translates into a business. Map your COGS model. Draw a line between upstream performance and downstream reality. Mark, in hard numbers, how much of the cost sits on each side of that line.
If downstream processing is a material cost or performance constraint in your process model, investigate the following priorities:
First: assess defensibility across strain, process, product, customer relationships and know-how, rather than assuming it resides in one part of the system.
Choose the next technical hire or partner against the measured constraint and the skills already available.
Second: compare downstream simplification with upstream improvement in the same cost model.
Compare the expected benefit and cost of improving titre, recovery or equipment utilisation. The preferred intervention depends on the model and the measured constraints.
Third: compare your team’s downstream experience with the work needed at the target scale. A manufacturing or development partner may fill a gap; confirm compatibility, knowledge transfer and commercial terms.
Treat downstream processing as part of product and process development from the beginning. Upstream and downstream performance must work together to produce an ingredient at the required specification and cost.
If titre doubled tomorrow, which cost and performance constraints would remain? Use the answer to decide what to measure next.
Sources and scope
This is an edited adaptation of the author’s full Substack essay, first published 2026-04-09. Examples and editorial frameworks are distinguished from measured evidence. Source checks and editing: 16 October 2026.
Read the original Substack version. Read the FoodEdge editorial standards or suggest a correction.
Cover photograph: Photo by Hans Reniers / Unsplash. Context image, not a photograph of a company or experiment discussed here. Photo licence.
